Enhancing the Prediction of Dam Deformations: A Novel Data-Driven Approach
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper proposes a data-driven method by conducting extensive comparisons, such as linear, arimax, adaboost, etc. Both plumbing data and InSAR data have been employed, in which the InSAR time series provides additional constraints to the deformation prediction. In general, this paper is well-written and provides interesting results to the community. I recommend this paper with minor revisions. Please see a few suggestions as detailed below.
Section 2.2: Did you use all the PS points from BBD or just select the key point of the dam structure?
Section 2.2: A description and analysis of radar geometry (incidence and azimuth angles) is important if the target is an arch dam.
Line 204: The model parameters are as big as 20000, how about the fitness in search space for different parameters listed in Table 2. Does there exist any over fitting?
Section 4.2: How to quantify the level of signal-to-noise ratio in PS points?
Figure 3. The axis label is too small, please make it clear.
Discussion: I think the author can also discuss the DS method for future work as it provides more dense scatterers on the dam structure.
Reference: I feel that authors can add some references for the arc dam or other types of dam deformation monitoring using InSAR. I remember that Pietro Milillo did some good work on this topic.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe manuscript proposes a new method for dam deformation prediction based on a data-driven approach that combines InSAR technology with traditional monitoring data, such as plumb systems, water levels, and temperature data, to improve prediction accuracy. The results show that the high spatial resolution of PS-InSAR data compensates for the limitations of traditional methods, demonstrating the advantages of data-driven approaches in improving prediction accuracy. The study recommends incorporating PS-InSAR data into dam monitoring systems to enhance both the comprehensiveness and accuracy of deformation predictions, especially for dams lacking traditional monitoring equipment. This work could be accepted after revisions. Below are some questions and suggestions for improvement:
1.The introduction needs to be expanded to demonstrate sufficient engagement with recent advancements in the field. The paper lacks enough citations regarding the use of InSAR for monitoring dam deformation. It is recommended that the authors include more references in this area to provide a clearer overview of the current state of research.
2.While the advantages of data-driven methods are emphasized, the manuscript does not address the computational requirements. The computational cost and training time of these methods should be discussed. It would be beneficial to include a comparison of the model's computational overhead to enhance the feasibility analysis.
3.The discussion section mentions the impact of the signal-to-noise ratio of PS-InSAR data, but no clear noise reduction method is provided. It is recommended that the authors suggest a noise reduction method to address this issue.
4.The conclusion does not clearly summarize the synergistic effect of PS-InSAR and plumb data. It generally mentions that "the combined use is recommended," but a clearer summary is needed regarding how these data sources interact and enhance prediction accuracy.
5.Both the abstract and conclusion sections are qualitative and lack specific, quantitative descriptions of the experimental results and data. The authors should include more detailed numerical descriptions to strengthen these sections.
6.In line 51, the use of the word "first" (e.g., “Importantly, to the best of our knowledge, this study presents the first fully data-driven prediction of dam deformations…”) should be avoided, as it could weaken the manuscript's persuasiveness. The authors should rephrase this to avoid overstatement.
7.It is recommended to add a spatio-temporal baseline figure of the Sentinel-1 data used in Section 2. This would provide a clear visual representation of the data's quantity, density, and baseline for the experimental setup.
8.The criteria for model selection mentioned in line 204 (e.g., “In total, we evaluated over 20,000 models for each time series”) are unclear. Please specify the detailed model search space and the selection criteria to help readers understand the decision-making process.
9.In Table 4, all points except for #10905 and #10961 were decomposed. Please provide an explanation as to why these two points were not processed.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf